Flex Image Edge Detect ⚡🅡🅞🅣🅘

This is Canny edge detection - the same algorithm behind a lot of ControlNet preprocessing - wrapped as a Flex node so its sensitivity can be driven by a reactive Feature rather than staying fixed. Use it as a stylistic effect (turning footage into a moving line-art/wireframe look) rather than for ControlNet conditioning; there are dedicated preprocessor nodes for that job elsewhere in the ecosystem, and this one's home is FlexFeatures/Targets/Images, squarely in the effects family.

How it works

Canny edge detection works with two thresholds rather than one: low_threshold and high_threshold (both 0–255, defaults 100/200) define hysteresis bounds. A pixel gradient above high_threshold is a definite edge; below low_threshold it's definitely not; anything in between only counts as an edge if it connects to a pixel that's already above the high threshold. That's why the two numbers exist instead of one simple sensitivity slider - it's how Canny avoids both missing faint-but-real edges and drowning the image in noise from every minor gradient.

feature_param lets a connected feature modulate either low_threshold or high_threshold (or None to disable modulation). feature_mode picks relative (nudges the base threshold up/down) or absolute (sets it directly from the feature), feature_threshold gates when that modulation activates, and strength scales how far it's allowed to swing the threshold.

Inputs and outputs that matter

• images - required.
• low_threshold / high_threshold - the whole effect lives here. Raise both together for a cleaner, sparser edge map; lower both for a noisier, denser one. The gap between them matters too - a wide gap picks up more of the "maybe" edges that connect to strong ones.
• opt_feature (optional) - for animating sensitivity over time, e.g. edges getting denser as audio intensity rises.

Output is a single IMAGE - the edge map, typically black background with white/bright edge lines.

How to install it

Via ComfyUI Manager, search "RyanOnTheInside." Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ryanontheinside/ComfyUI_RyanOnTheInside
cd ComfyUI_RyanOnTheInside
pip install -r requirements.txt

then restart. Edge detection runs through OpenCV, which is part of the pack's shared requirements - no separate download needed if the install completed cleanly.

Common issues & troubleshooting

Output is nearly all noise, or nearly blank. This is almost always the threshold pair, not the image. Too-low thresholds pick up every minor gradient as an edge (noisy, chaotic output); too-high thresholds miss everything but the strongest contrast lines (a nearly empty frame). Start near the defaults (100/200) and adjust in both directions to see where your source image's real structure lives.

Result looks different frame to frame even on a static-looking clip. Canny is sensitive to per-frame noise (compression artifacts, grain, slight exposure shifts), so a visually "static" video can still produce flickering edge maps. If you need temporal stability, consider pre-smoothing the source or blending consecutive edge-detected frames downstream - this node itself has no temporal smoothing of its own.

Feature modulation on thresholds produces a broken-looking edge map at extremes. Remember high_threshold needs to stay meaningfully above low_threshold for Canny's hysteresis logic to behave sensibly - if a feature pushes one threshold past the other, you can get inverted or degenerate-looking results. Keep an eye on the effective range your feature modulation covers, not just its nominal min/max.